Metrological evaluation of neural radiance fields for quality control of mechanical and additive manufacturing components: a multi-material comparison against professional optical scanning

Abstract

This study provides a systematic metrological evaluation of Neural Radiance Fields (NeRF) against professional laser-light and structured light scanning across three specimen types: a polymer (PLA) bracket, a metallic bracket, and a matte ornamental owl. Consumer-grade smartphone and DSLR cameras were benchmarked against professional scanning as metrological ground truth. On bracket specimens, NeRF achieved RMS deviations of 1.48 mm (metal bracket) and 1.32 mm (PLA bracket). On the ornamental owl, DSLR-based NeRF achieved RMS 0.27 mm vs. 0.41 mm for smartphone (a 34% reduction) confirming input image quality as the primary accuracy lever. Repeatability confirmed CV below 6% for the metal bracket (n = 3 trials) and CV = 2.1% for the DSLR owl reconstruction. NeRF is suitable for non-critical quality screening across diverse geometries while not yet reaching sub-100 µm metrology-grade precision. To the best of the authors’ knowledge, this is one of the first multi-material, multi-geometry NeRF evaluations conducted under standardised metrological conditions.